Hard exudate based severity assessment of diabetic macular edema from retinal fundus images

Abstract

Diabetic macular edema (DME) is a consequence of diabetic retinopathy characterised by the abnormal accumulation of fluid and protein deposit in the macula region of the retina. Prior disclosure of even a trivial trace of DME is essential as it could consequently lead to blurred vision. DME can be diagnosed by the presence of exudates (glossy lesions) in the retinal fundus images. In this work, OD and macula are detected using morphological operation and hard exudates are segmented. Exudates are classified using early treatment diabetic retinopathy standard as normal, moderate or severe cases. The proposed work also incorporates the extraction of various features from the retinal fundus image. Various textural and exudate features are extracted and fed to a classifier to detect DME. Experiments are performed on a publically available database. Performance is evaluated with metrics like accuracy, sensitivity, specificity and accuracy. The results obtained are promising.